The mathematical prediction model for shrinkage of pavement cement concrete under multi factors is difficult to establish. Therefore, the BP neural network model was developed to predict shrinkage of concrete. Results show that BP neural network can accurately predict the shrinkage of concrete and the model has good ability to generalize. By comparing five different algorithms, the Trainlm algorithm is quick to be trained but has big error, whereas the Traingda algorithm can be trained not as quick as Trainlm but has the minimum error. Hence, the neural network model by applying Traingda algorithm can well reflect the nonlinear relationship between materials mix proportion and the dry shrinkage ratio of pavement cement concrete.